[Beowulf] Athlon64 / Opteron test
Robert G. Brown
rgb at phy.duke.edu
Fri May 14 08:49:34 PDT 2004
On Fri, 14 May 2004, James Cownie wrote:
> > I think they'd still have trouble maintaining a market, because
> > Opterons are relatively easy to port to and will in principle run
> > i386 code (badly, of course) native.
> The "(badly, of course)" should definitely be removed.
> The mass market for the AMD64 chips at the moment is gamers running
> that other OS which so far _only_ runs in 32 bit mode.
> The gamers like the AMD64 chips because they're the fastest 32 bit x86
> chips you can buy.
> (Check any of the benchmark sites for proof.)
> You may have been confusing the AMD64 bit chips with the Itanic, which
> does have a very slow 32 bit x86 mode.
Actually, I was referring to the following:
rgb at ganesh|B:1117>uname -a
Linux ganesh 2.4.20-30.9 #1 Wed Feb 4 20:45:39 EST 2004 i686 athlon i386 GNU/Linux
rgb at ganesh|B:1118>pwd
rgb at ganesh|B:1119>make
make: Nothing to be done for `all'.
rgb at ganesh|B:1121>ssh s02
rgb at s02|B:1001>cd Src/Ospin
rgb at s02|B:1002>uname -a
Linux s02 2.4.22-1.2188.nptlsmp #1 SMP Wed Apr 21 20:12:16 EDT 2004 x86_64 x86_64 x86_64 GNU/Linux
rgb at s02|B:1003>./Ospin
-bash: ./Ospin: /lib/ld-linux.so.2: bad ELF interpreter: No such file or directory
rgb at s02|B:1004>
Where on a regular Athlon (basically RH9 i386) I compile Ospin (my
current application) and then I ssh to s02 (fedora x86_64) to run the
binary, it barfs. Probably missing a compatibility library, maybe the
library is around somewhere but not installed but I don't care much
...it also refers to anecdotal accounts that numerical performance
significantly degrades if one runs i386 code compared to recompiled
x86_64 code. That may leave it plenty fast enough for gamers (who may
be running memory bus bound vector float transformation code a lot and
hence get benefit from the fast memory without getting much advantage
from the CPU), but still slower than the optimum one would like running
numerical applications that take hours to days to complete.
So OK, "badly, of course" is perhaps overly harsh. "Better than
Itaniums" would be fair, I think. So would "more slowly than native 64
bit recompiles" which is what most people care about anyway.
I stand corrected;-)
Robert G. Brown http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567 Fax: 919-660-2525 email:rgb at phy.duke.edu
More information about the Beowulf